Recent Commits & Progress

Yesterday was a record 24-commit day. Since the evening retrospective, 10 additional commits landed:

• 2aced04a fix: log warn for non-NotFound I/O errors in load_dedup_file (#2258)
• 8f3dd0eb bug: start_run and complete_run audit trail silently drops all DB errors (#2259)
• 59ea737a bug: get_mentions silently drops mentions on parent-issue fetch failure — no observability (#2257)
• 35cdd008 refactor: move NoopBackend test helper to dedicated test_helpers.rs (#2252)
• b25c3d11 bug: parse_success_output only checks result field — misses AgentResponse JSON in earlier NDJSON messages (#2251)
• a0728c14 bug: write block_reason before blocking in review.rs and review_poll (#2250)
• 24daca1f bug: set_cooldown_async returns true despite KV write failure — violates contract (#2249)
• 76c67939 bug: wait_for_cooldown returns misleading error when agents are degraded (#2248)
• c9debae0 test: fix router_round_robin_routes_task broken by model_map change
• 5c3a9770 bug: mergeability deferral resets approved PR to NeedsReview (#2247)
• 8fd56391 bug: next_round_robin_agent skips cooldown check but ignores degraded state (#2246)
• 6fdf5377 bug: handle_review_changes writes block_reason twice with inconsistent data (#2245)

The reliability push has now been running for two full days with no sign of slowing. Focus remains on correctness: audit trail gaps, misleading error contracts, silent data drops at DB and observability boundaries.

Operational Health

Overall: healthy. Pipeline is processing work, review automation is functioning, no blocked tasks visible. One minor structural concern: CLI/service version drift.

Live concerns

1. CLI/service version mismatch

   CLI:     0.60.123
   Service: 0.60.131  ✗ mismatch (8 versions behind)

   The evening retro flagged this as a priority but upgrade wasn't confirmed. With 10 more commits since then, the gap is now 8 versions. Run:

   brew upgrade orch && brew services restart orch && orch version

2. #2254 in review, waiting for CI

   Task 2254 (bug: next_round_robin_agent ignores model availability) went through a full automated review cycle this morning:

   • First review (opencode/nemotron) → RequestChanges (fallback branches still bypass model availability check)
   • Agent re-dispatched (opencode/minimax-m2.5-free) → completed fix, pushed to PR #2260
   • Second review attempt (opencode/nemotron) → "Provider returned error" → reset to needs_review
   • Third review (minimax/opus) → Approved
   • Status: in_review, CI pending (1 of 2 checks passing at last check)

   The automated loop is working. No human intervention needed — just waiting for CI to go green and auto-merge to fire.

What looks healthy

• No blocked tasks. Only 1 external task in the queue (2254, actively progressing).
• No active cooldowns. KV cooldown table is empty — all agents are routable.
• Codex is fully recovered. Yesterday's retro noted the cooldown expires Apr 9. This morning: 40 successes in 24h, confirming full recovery.
• Router LLM pool exhaustion (#2183) not observed. No "router pool exhausted" log entries this morning. Yesterday's #2222 (router LLM skips degraded agents) appears to have reduced the trigger frequency, or load was simply lower this morning.
• Review automation is end-to-end functional. Three successive review agent invocations on #2254 this morning — one failed ("Provider returned error"), retried cleanly with a different agent, approved.

24h run outcomes

Notes:

• olm (gemma4) appears in run stats for the first time — a new agent/model is being exercised.
• qwen3.6-plus-free failures are down to 2 (from 10 yesterday) — yesterday's #2228 (ModelUnavailable classification) appears effective.
• kimi rate limits (4) are minor; exponential backoff is handling them generically.
• opencode/nemotron-3-super-free showed "Provider returned error" in review (handled by retry), but also 8 successes in other runs.

Last 12h task activity

Error volume (84 in 12h) is similar to yesterday and consistent with a high-throughput pipeline where transient failures are expected and retried.

Retro Follow-Ups

Priorities for Today

1. Upgrade CLI/service — 8-version gap. Run:

   brew upgrade orch && brew services restart orch && orch version

2. Wait for #2254 to merge — already approved, CI pending. No action needed unless CI fails or auto-merge stalls.

3. Investigate olm agent — appears in run stats for the first time with gemma4. Three failures, two successes. If this is a new agent integration being tested, watch for failure patterns.

4. Confirm router LLM exhaustion stays quiet — no events this morning, but check again after the service upgrade in case behavior changes.

5. Watch kimi rate limits — 4 rate limit events in 24h is minor, but worth monitoring to confirm the exponential backoff is preventing cascading failures.

← All updates